Method for characterizing a tubular object such as a blood vessel by ultrasound imaging

Abstract

Method for characterizing a tubular object by ultrasound imaging, the method comprising the steps of: - Acquiring (201) several ultrasound signals originating from the reflection of an ultrasound field emitted by the transducer on an area of ​​interest in a cross-sectional plane of the object, for different positions of the transducer relative to said area, all of the ultrasound signals forming an ultrasound image of the area, - for each signal corresponding to a vector of the image, applying (202) a first predetermined filter to the signal, - Selecting (203), from all of the signals, the signal for which the result of the filter has the highest absolute value extremum and recording the abscissa of this extremum, - Determining (204) the center of the object from the recorded abscissa and the speed of the ultrasound signal.

Description

[0001] The invention relates to the field of ultrasound imaging, in particular medical imaging, and more specifically relates to a method for determining the center and diameter of a tubular object such as a blood vessel.

[0002] Ultrasound imaging, particularly in the medical field, aims to image different types of objects or organs in the human body to better characterize them.

[0003] In the field of medical imaging, there is a need to characterize blood vessels, particularly arteries. Indeed, arteries are a true health indicator, as they provide essential information on the state of health of the human body. Real-time monitoring of arteries allows the tracking of biological markers such as blood pressure to better understand the functioning of the human body and anticipate the development of diseases. The development of ultrasound-based technologies allows the design of compact imaging systems that offer a non-invasive solution. Real-time monitoring requires that the entire measurement chain, from acquisition and artery location to the extraction of relevant information, must be fully automated.

[0004] In this general area, there is a particular need to characterize arteries via the detection of their center and the measurement of their diameter.

[0005] Ultrasound imaging methods that offer solutions for characterizing blood vessels most often focus on the longitudinal dimension of the vessel, for example the carotid artery.

[0006] These methods are diverse but can be grouped into three non-exhaustive categories which include manual, semi-automatic and automatic methods.

[0007] Manual methods rely on human expertise to manually determine the position of arteries, for example using graphical tools applied to ultrasound images.

[0008] In the case of semi-automatic methods, assistance algorithms work in collaboration with a human. For example, the algorithm can continue to automatically locate the artery based on the human as a guide, who validates, corrects, and provides reference information to help the algorithm find the correct location.

[0009] In the fully automatic methods category, the algorithm automatically determines the artery location. Most of these automatic methods specifically address the problem of carotid artery location and do not work optimally with other arteries. Indeed, the carotid artery is a less challenging case than other arteries such as the radial artery for several reasons. The carotid artery is larger with thicker walls and has a different intensity profile than the radial artery, with less noise and more contrast. Because the diameter of the carotid artery is larger than that of the radial artery, ultrasound images have better contrast. This makes it easier to locate the center of the carotid artery simply by searching for the area with the lowest intensity profile.Compared to the radial artery, ultrasound images of the carotid artery naturally present fewer anatomical objects that can be confused with the artery. The larger size of the carotid artery also contributes, as it allows for limiting the fields of ultrasound images, which allows for images that contain the fewest noisy objects. Most methods developed for the carotid artery focus on the longitudinal dimension, which further facilitates localization by simply identifying the horizontal trace with the lowest intensity profile. Other anatomical objects have a less noisy footprint on this dimension. Most algorithms localize the area with the lowest intensity vertically, scanning the entire image; subsequently, the artery is localized according to a consensus of the identified points after removing outliers.These longitudinal dimension-based methods are sensitive to probe orientation and all require careful positioning of the probe on the skin during acquisition. Orientation uncertainty will make the measurement of diameter and other properties inaccurate. This problem can be circumvented by using multidimensional probes, but these are more expensive.

[0010] Although the transverse dimension is more difficult to process, because the shape of anatomical objects on this dimension appears frontally and disrupts localization, particularly in the case of the radial artery, the advantage of the transverse dimension is that it does not require precise positioning of the probe on the skin, and the diameter measurements remain robust and reliable with respect to probe orientation errors.

[0011] Existing solutions for automatic localization on the transverse dimension of arteries or more generally blood vessels are rare and all suffer from one or more limitations. Methods based on deep learning are effective but consume a lot of energy, are not suitable for small portable systems, they are complex to implement and require a lot of resources. It is difficult to achieve real-time performance, especially when it comes to a portable computer. These methods are found integrated into assistance interfaces for which the requirements are not limited in time and resources. Similarly, neural networks, characterized by their black box aspect, suffer from the problem of unpredictability. It has been shown that a single pixel of an image can lead a neural network to provide aberrant data.This problem is also the main cause slowing down the adoption of artificial intelligence in the medical field, and in areas with serious consequences.

[0012] Doppler-based solutions allow localization through the frequency change related to blood flow. These methods are limited to vertical localization, where the depth of the artery is located. Horizontal localization is performed manually. Another disadvantage of these methods is that Doppler localization requires monitoring ultrasound frequencies over time, which means that all these methods require the time dimension to locate the artery.

[0013] Other limitations are evident in several aspects, such as speed of execution and meeting real-time standards. Iterative search algorithms, for example, are slow and require considerable resources to achieve real-time operation.

[0014] Algorithms requiring a reference model and prior information are a handicap. Another aspect is related to algorithms that are sensitive to noise and require preprocessing to condition the received data before executing the localization method. Some existing solutions rely on landmarks such as contours, which makes them vulnerable to data imperfections. These methods often rely on preprocessing, consensus, and outlier removal algorithms to ensure robustness, which adds complexity and computation.

[0015] Other methods are applied to other technological solutions that go beyond the scope of ultrasound, such as the so-called intravascular ultrasound or IVUS methods and the so-called optical coherence tomography or OCT methods. Although these methods process arterial images, they do not address the same problem, because the profile of the processed data is different, in terms of intensity, noise and difficulties. Also, these methods inherit the disadvantages of these technologies such as invasiveness in the case of IVUS methods and depth limitation in the case of non-invasive OCT methods.

[0016] Patent application WO2016057233 presents a method for estimating a non-invasive continuous blood pressure waveform using ultrasound and an automated cuff. It belongs to the category of non-invasive continuous blood pressure measurements. The proposed system measures physical characteristics such as the geometry, elasticity, and deformation of a blood vessel, as well as other external physical parameters. Computer modeling and processing of the measured signals are used during cuff inflation and / or deflation to iteratively estimate the subject's blood pressure. The ultrasound part allows estimating the diameter of the artery and calculating its elasticity.

[0017] The artery localization in this solution is limited by several points: first, the transducer arrays are fixed by a skilled operator at an approximate location and the artery is localized in three dimensions by an iterative search method of the assumed geometric characteristics and ultrasonic response. The search is not only iterative, which is computationally intensive, but it is done in a 3D search space, therefore on a large amount of data. This limits the execution speed especially for devices limited in resources and energy consumption. In addition, this search in the 3D space is performed by comparing at each iteration, with a reference artery model built through personal prior information about the subject, such as age, height, etc.In addition to requiring a reference model and prior information, this solution also requires a two-dimensional probe to enable acquisition in the 3D space in which the iterative search algorithm runs. With slow localization, refreshing the artery position imposes a trade-off for speed, accuracy, and the person's movement.

[0018] The document [1] presents a method for characterizing the carotid artery that can be applied to the longitudinal and transverse dimension of this artery. It consists of a contour detection algorithm based on the "Snake" method and equipped with a mathematical artery reference model in the form of B-splines that deforms in scale, translation and rotation. The method converges to the carotid artery contours using an optimization algorithm. The convergence criterion is ensured by a sum of the absolute differences between the model and the image at each point of the image. The model parameters are optimized iteratively using the energy minimization algorithm, then the sum of the absolute differences is calculated at each point. The circular shape of the reference model is used to simplify the calculation. A gradient-based optimization algorithm is applied to refine an accurate segmentation of the detected artery contours.

[0019] In this document, the shape of the carotid artery is quite large and circular, and the thickness of its wall is quite large, which makes it easy to distinguish the contours of the layers, on which the algorithm is based. This solution is not applicable to other types of arteries such as the radial artery for several reasons. First of all, the radial artery often appears with a shape that is neither perfectly circular nor perfectly rectangular, because of its small size. The wall of the radial artery is small, and does not allow its layers to be distinguished with standard probes at low central frequencies, this requires high-quality probes at higher ultrasound frequencies.For this reason and since the method in reference [1] relies on the differences in local contrast between the inner and outer contours, it will be difficult for the algorithm to accurately locate the region of interest in cases where the artery or surrounding tissue lacks texture or contrast, which is the case of the radial artery.

[0020] Also, anatomical objects on transverse images of the carotid artery, such as the jugular vein, have non-circular shapes and are easily discriminable from the carotid artery. For this reason, the algorithm is not likely to converge on other objects. In the case of the radial artery, several anatomical objects have a shape and positioning very similar to the radial artery, which means that the method is likely to converge on the other objects instead of the radial artery, because the criterion of the sum of absolute differences will be satisfied also, for example, by the veins neighboring the artery.

[0021] The method requires image enhancement preprocessing based on adaptive histograms to reduce noise. This is because noise can significantly increase the convergence time of the optimization algorithms on which the method is based. The convergence time is variable and depends on the intensity profiles of the acquired image, its complexity, the position of the artery, and its similarity to the model. It also depends on the initialization parameters integrated into the model. If the image contains complex anatomical structures or if the artery itself has irregular shapes, the comparison approach with the reference model may not work well within the time limits.

[0022] Also, the model may not be flexible enough to handle these variations, resulting in poor alignment. Therefore, this method is more suitable for applications that do not require strict real-time constraints, and regular time-of-flight measurements at fixed intervals. The method compares image pixels with a model at each point of the image while calculating the absolute sum of differences criterion over the pixels each time. This operation is computationally intensive, which requires the method to use optimization algorithms to reduce the computation time. The disadvantage of this technique is that it makes the solution entirely dependent on the convergence time of the optimization algorithm, which is an iterative algorithm dependent on the complexity of the image and vulnerable to convergence problems such as local minima.Additionally, optimization may converge on veins rather than artery on radial artery cross-sectional scans.

[0023] The method presented in [2] concerns the localization of the carotid artery in transverse images. The method is based on the circular Hough transform, a method that allows detecting lines in the image and has been adapted for detecting circles. The method processes sequences of 5 to 15 consecutive images. Each image is preprocessed to optimize brightness and contrast. Then, a strong Gaussian filter reduces noise. Although several fine morphological structures have been destroyed in this process, the carotid artery appears brighter. Then, the Hough transform detects all dark circles by considering all radii that fall within a reasonable range. The coordinates of the detected centers (xi, yi) and radii (ri) are collected. For each of the detected dark circles, the brightness values ​​of their pixels are evaluated.The circle with the darkest values, i.e., the minimum brightness, is elected as the "candidate circle," and its center coordinates (xci, yci) with a radius (rci) are recorded in a matrix of potential candidates. Once all B-mode images have been processed, the matrix contains the results obtained from each image. The most frequent triplet (xc, yc, rc) appearing in the matrix is ​​selected as the final choice.

[0024] This third method is also not suitable for the radial artery for the following reasons. The carotid artery has a fairly circular shape, it stands out as a fairly dark area and is easy to discriminate from other anatomical objects. The shape of the radial artery on ultrasound images does not appear perfectly circular. In addition, there are several anatomical objects similar to the artery, for example, veins, which are nearby, sometimes darker than the radial artery and also with a circular shape. The Hough transform can be adapted to other more complex shapes to apply it to the radial artery, moreover, this considerably increases the intensity of the calculations, because the circular shape as presented in the document [2] already performs a search in a 3D space which requires a much more intense calculation than with the standard Hough transform.Indeed, for each circle diameter, the method performs the calculation of a new accumulator on the whole image. The radial artery can be easily deformed, for example towards an ellipse shape which implies an additional parameter to be taken into account by the Hough transform. The shape of the radial artery is much more deformed than that of the carotid artery, because the size is smaller which makes the contours smaller and therefore more deformed by the noise observed on the ultrasound images. The Hough transform provides multiple detections and the choice according to the darkest area will not work in the case of the radial artery, because the neighboring veins can be darker. These multiple detections cause a large number of false detections, which requires either the use of more complex criteria or to consider the temporal dimension in order to improve the robustness of localization.For this reason, the method requires 5 successive images. The circular Hough transform requires an edge detection step and thresholding adjustments. Edge detection is vulnerable to noise in the image, which implies the need for preprocessing for enhancement and noise filtering. For this reason, the method in the paper specifies that strong Gaussian filtering is necessary.

[0025] All the above-mentioned methods have the disadvantages described. There is then a need for a new method which is applicable to the characterization of the cross-section of an artery, in particular the radial artery, which does not require a two-dimensional probe and which is adapted to the acquisition environment of the artery to be characterized, in particular noise and the presence of other biological objects.

[0026] The invention allows automatic localization and characterization of a radial artery in its cross-section. It involves a first phase of localization of the center of the artery from several ultrasound acquisitions forming a 2D image of the cross-section including the section of the artery, then a second phase of determination of the diameter of the artery from the ultrasound measurement corresponding to the path passing through the center of the artery.

[0027] The invention does not require precise positioning of the probe on the skin by an expert, and tolerates angular positioning uncertainties. This allows for reliable measurements even in the presence of movement of the subject and the probe relative to the subject. The diameter measurement is made at the most relevant horizontal point in the center of the artery, which is the point that provides the most accurate diameter measurement.

[0028] The principle of the invention is based on a convolution filter approach and includes a basic filter specifically dedicated to detecting an artery or more generally a tubular object. Once the center of the artery is located, methods for extracting the properties of the artery are applied locally to the filter result.

[0029] The proposed method has low complexity and can be implemented easily, as almost all localization operations are performed as single operations, which is a considerable simplicity advantage and allows the method to be embedded on resource-limited portable systems and to perform time-of-flight measurements. The proposed method does not require any comparison with a reference model, nor calculation of a resemblance criterion at each point, nor any intensive iterative calculation, nor risk of iterative convergence and local problems that optimization algorithms suffer from. Another advantage of the invention is that it can be executed in synchronization with the acquisition in an incremental manner as signals are received, without the need to wait for the end of the acquisition. This allows a rapid refresh of the artery location.Unlike existing methods, the proposed method does not need to store a large number of parameters, as it gradually updates the summation and the coordinates of the extremum point as the computation progresses. Some existing solutions rely on landmarks such as contours, which makes them vulnerable. Some of these solutions use noise enhancement or outlier removal methods, which increases complexity and computation. The proposed solution is based on global information, which makes it robust against noise and does not require image enhancement preprocessing or outlier removal algorithms.

[0030] The proposed solution performs localization directly on the spatial dimension of a single transverse acquisition and does not require multiple acquisitions on the temporal dimension, unlike Doppler-based methods. The proposed method does not require prior information about the person, nor a reference artery model. With fast localization, the artery position is refreshed at high speed and offers the possibility of having a higher measurement frequency, more precision and more freedom of movement of the person.

[0031] Although the invention is described in the particular context of the characterization of a blood vessel for medical imaging applications, it can also be applied in the field of non-destructive testing by ultrasound to characterize any type of tubular-shaped object.

[0032] The subject of the invention is a method for characterizing a tubular-shaped object by ultrasound imaging, the method comprising the steps of: Acquire by means of an ultrasonic transducer, several ultrasonic signals originating from the reflection of an ultrasonic field emitted by the transducer on an area of ​​interest in a cross-sectional plane of the object, for different positions of the transducer relative to said area, all of the ultrasonic signals forming an ultrasonic image of the area, Choose a dimension of the ultrasonic image and for each signal corresponding to a vector of the image according to the chosen dimension, apply a first predetermined filter to the signal, the filter being configured so as to transform a first signal comprising two extrema of the same sign into a second signal comprising an extremum of opposite sign located between the two extrema of the first signal, Select, from all of the signals, the signal for which the result of the filter has the extremum of the highest absolute value and note the abscissa of this extremum,Determine the center of the object from the abscissa taken and the speed of the ultrasonic signal.

[0033] According to a particular aspect of the invention, the filter is applied to the envelope of the ultrasonic signal or to the absolute value of the ultrasonic signal.

[0034] According to a particular aspect of the invention, the filter is applied to the signal over a sliding window of predefined size depending on the size of the signal and / or a priori information on the dimension of the object, the filter being defined over at least three consecutive time intervals by three respective functions each weighted by a coefficient, the coefficients associated with two consecutive time intervals being of opposite signs.

[0035] According to a particular aspect of the invention, the dimension of the second time interval is chosen so as to be strictly less than the minimum diameter of the object to be characterized.

[0036] According to a particular aspect of the invention, the filter is defined over at least two additional time intervals.

[0037] According to a particular aspect of the invention, each of the functions is taken from: a sum, a maximum value, an average or a combination of these functions.

[0038] In an alternative embodiment, the method according to the invention further comprises the steps of: Select the acquired ultrasonic signal for which the center of the object has been determined, Apply a predetermined threshold to said selected ultrasonic signal, Detect at least two extrema of said signal greater than the threshold, Select the pair of extrema, comprising a first extremum and a second extremum, closest to the center of the object and located on either side of the center of the object, note their respective time abscissas and deduce the internal diameter of the object from the difference between the two abscissas and the speed of the ultrasonic signal.

[0039] In an alternative embodiment, the method according to the invention comprises the steps of: Select a third extremum greater than the threshold and located immediately before the first extremum, Select a fourth extremum greater than the threshold and located immediately after the second extremum, Record the time abscissas of the third extremum and the fourth extremum and deduce the external diameter of the object from the difference between the two abscissas and the speed of the ultrasonic signal.

[0040] According to a particular aspect of the invention, the object is a blood vessel, for example an artery.

[0041] According to a particular aspect of the invention, the step of acquiring several ultrasonic signals comprises the sub-steps of: Position a transducer comprising several aligned elements, on an area of ​​the skin so as to image a cross-section of the blood vessel, Carry out several successive ultrasound acquisitions from different emission points located on the alignment axis of the elements, each ultrasound emission being carried out in a direction substantially perpendicular to the alignment axis.

[0042] The invention also relates to an ultrasound imaging device comprising an ultrasound transducer and a processing unit configured to carry out the steps of the method according to the invention.

[0043] Other features and advantages of the present invention will become more apparent upon reading the following description in relation to the following appended drawings. [ Fig. 1 ] represents a diagram of a multi-element ultrasonic probe capable of carrying out an ultrasonic signal acquisition sequence to image a blood vessel, [ Fig. 2 ] represents a flowchart describing the steps of implementing a method for detecting the center of a blood vessel according to an embodiment of the invention, [ Fig. 3a ] represents an ultrasound image of an area of ​​a transverse plane including a radial artery, [ Fig. 3b ] represents the image of the figure 3a filtered by means of a first filter defined according to the invention, [ Fig. 3c ] represents an image of the figure 3b filtered by means of a second filter defined according to the invention, so as to allow the detection of the radial artery, [ Fig. 4a ] represents a time diagram of an ultrasonic signal acquired by means of the probe of the figure 1 , [ Fig. 4b ] represents an example of a filter intended to be applied to the signal of the figure 4a , [ Fig. 4c ] represents the result of applying the filter of the figure 4b at the signal of the figure 4a [ Fig. 4d ] represents the result of an additional filtering step applied to the signal of the figure 4c , [ Fig. 5 ] represents a flowchart describing the steps of implementing a method for determining the diameter of a blood vessel according to one embodiment of the invention, [ Fig. 6 ] represents an example of an ultrasound signal used to determine the diameter of a blood vessel using the method of figure 5 ,

[0044] There figure 2 represents, on a flowchart, the main steps of implementing a method for determining the center of a blood vessel, for example a radial artery, according to one embodiment of the invention.

[0045] The method begins at step 201 with an acquisition of ultrasound signals in a transverse plane of the artery or vessel.

[0046] This acquisition is carried out using a probe shown in the figure 1 .

[0047] The acquisition probe is advantageously a linear multi-element probe 101, i.e. one which comprises several ultrasonic elements 102, for example piezoelectric elements, aligned in a row. Alternatively, a two-dimensional probe, i.e. one comprising a matrix of ultrasonic elements, can also be used.

[0048] In the example of the figure 1 , the probe 101 is positioned in contact with the skin 103 so that the row of elements 102 is located substantially in a transverse plane of an artery 104 to be imaged. The probe 101 is not necessarily centered on the artery 104, it is sufficient that the artery is covered in emission and reception by the ultrasound beam generated by the probe so as to obtain a transverse imprint of the artery.

[0049] The acquisition 101 is carried out by means of a successive scan during which, at each step, an ultrasonic beam is emitted by a group of elements 102 comprising at least one element, in a direction perpendicular to the axis of the alignment of the elements. The same elements are active in reception to generate an ultrasonic signal corresponding to a predefined acquisition duration and to an axis substantially perpendicular to the group of elements 102.

[0050] This step allows an ultrasound signal to be acquired. It is then iterated by shifting, according to a sliding window, the active ultrasound elements by one element and then performing a new acquisition. Thus, by performing several successive acquisitions using a group of elements of fixed size which scans all the elements of the probe, several ultrasound signals are obtained which together form an ultrasound image of a transverse plane of the artery 104.

[0051] This acquisition makes it possible to obtain in two dimensions a transverse ultrasound impression of the targeted radial artery as well as any other anatomical objects present around it such as veins.

[0052] Without departing from the scope of the invention, other ultrasound acquisition methods may be envisaged insofar as they allow a 2D image of said transverse plane to be obtained. On the image obtained, which corresponds to a matrix of signal samples, the vertical position represents the depth, i.e. the distance between the probe and an anatomical element located under the probe. The horizontal position represents a point on the skin 103 corresponding to the emission point of the ultrasound beam.

[0053] There figure 3a represents an example of an ultrasound image obtained with identification of the radial artery 104. It can be seen that in this image, the imprint of the artery 104 is difficult to detect because it is drowned in noise and polluted by the imprints of other anatomical objects.

[0054] An objective of the invention is to identify the point on the skin located directly above the center of the artery in order to determine the ultrasound measurement which corresponds to a path passing through this center.

[0055] There figure 4a represents an example of an ultrasound signal acquired during a single acquisition. The signal from the figure 4a corresponds to a column of the matrix of the figure 3a . In fact, each column of this matrix corresponds to an acquisition from a point on the surface of the skin. More precisely, the signal from the figure 4a corresponds to the envelope or absolute value of the acquired ultrasonic signal.

[0056] In step 202, one or more filtering steps specifically adapted to identify the center of the artery are then applied to each of the acquired signals (each column of the 2D matrix).

[0057] As can be seen on the figure 4a , an acquisition in the transverse plane of the artery is characterized by two peaks or groups of amplitude peaks which correspond to the echoes on the walls of the artery. Between these two peaks is the interior of the artery.

[0058] A first filter is applied to the signal of the figure 4a . This filter is defined so as to transform the signal of the figure 4a which includes two extrema of the same sign in another signal which includes an extremum of opposite sign located substantially halfway between the two extrema of the first signal.

[0059] An example of a filter is shown in figure 4b . This filter corresponds to a time slot. It has a total size greater than the diameter of the artery that we wish to detect and is composed of three successive parts. The central part 401 is set to a positive value, for example equal to 1, over a duration corresponding to a distance less than the internal diameter of the artery.

[0060] The other two parts 402, 403 are set to a negative value, for example equal to -1.

[0061] More generally it is possible to replace the value 1 with another positive value and the value -1 with another negative value.

[0062] By applying a convolution of this filter with the acquired signal, the outer parts of the filter 402, 403 contribute negatively to the convolution results, while the central area 401 contributes positively to the convolution results.

[0063] When this filter is applied to the signal acquired at the level of the area corresponding to the artery, the amplitude peaks corresponding to the walls of the artery will be aligned with the outer areas of the filter 402, 403, their contribution will therefore be summed negatively. Conversely, when the central part of the filter does not coincide with the central area between the two amplitude peaks but coincides with one of the amplitude peaks, then the result of the convolution with the filter will be a positive value.

[0064] There figure 4c shows the result of applying the filter of the figure 4b at the signal of the figure 4a . It can be seen that the obtained filtered signal presents positive amplitudes everywhere except in the area corresponding to the interior of the artery.

[0065] There figure 4d shows the final result obtained by setting all positive values ​​to zero. The extremum of the signal of the figure 4d corresponds to a point in the inner zone of the artery for which there are only negative contributions, i.e. a situation where the central zone 401 of the filter is applied between the two amplitude peaks corresponding to the walls of the artery.

[0066] Thus, by analyzing the filtered signals of the type of the figure 4d for all acquisitions, it is possible to locate the center of the artery.

[0067] The filter described in the figure 4b is a non-limiting example and can be adapted to the situation that one wishes to analyze. In particular, the dimensions of the three parts 401, 402, 403 of the filter can be adapted according to the maximum dimensions of the blood vessels that one wishes to detect.

[0068] In particular, the dimension of the central part 402 can be chosen to be strictly less than the minimum diameter of a vessel that one wishes to characterize. Alternatively, this dimension can be chosen to be of the order of the average diameter of the vessels to be characterized. The choice of dimensioning also depends on the type of basic function applied by the filter (maximum, minimum, average or sum in particular).

[0069] Other filter variants may be considered as described now.

[0070] If we note S the envelope or the absolute value of the acquired ultrasonic signal, this signal being composed of N x samples: S = i x 1 ≤ x ≤ N x , , where i(x) is the ultrasonic intensity (amplitude) at point x of the signal S .

[0071] We note ROI r ={i( a r ), i( a r + 1), i(a r + 2)..., i( b r ), |ar , ≤ x ≤ b r ; ( a r , b r ) ∈ S}, where, ROI r is a region of interest r represented by the set of points of the signal S located in the interval [ a r , b r ] . The filter is defined by applying functions to multiple regions of interest. In the example of the figure 4b the number of regions of interest is three but can be greater than three as will be described later.

[0072] The general form of the filter applied to the signal is given by: F = ∑ r = 1 N r α r f r ROI r , Or f r< is a strictly positive valued function applied to the signal, such as, for example, the maximum, the sum or the average over all the points belonging to the region ROI r . α r is a predefined weighting coefficient. In the example of the figure 4b , there are three areas of interest, so N r =3 and the functions f r< are all the sum function with coefficients equal to -1 for the outer areas of interest and +1 for the central area of ​​interest of the filter.

[0073] More generally, the coefficients can have an absolute value other than 1, the filter F is then written: F = α 1 f 1 ROI paroiProximale + α 2 f 2 ROI ParoiDistale − α 3 f 3 ROI V a l l é e

[0074] The ROI Valley area of ​​interest corresponds to the central area of ​​the filter and the ROI Proximal Wall and ROI Distal Wall areas of interest correspond to the outer areas of the filter.

[0075] For the example of the figure 4b , the filter can be expressed as: F = somme ROI V a l l é e − somme ROI paroiProximale − somme ROI ParoiDistale

[0076] Other forms of filter can be considered, either calculated directly or calculated locally on the points detected by a first filtering step.

[0077] In particular, filters composed of five areas of interest can be considered in order to take into account the external and internal walls of a vessel. Indeed, the external wall and the internal wall are separated by a non-echogenic middle layer. The acquired signal can therefore present, for each of the proximal and distal walls, two amplitude peaks separated by a zone close to zero, the two peaks corresponding to the external and internal walls. F = + α 1 f 1 ROI paroiProximale − α 4 f 4 ROI CoucheMedianProximale + α 2 f 2 ROI ParoiDistale − α 5 f 5 ROI CoucheMedianDistale − α 3 f 3 ROI V a l l é e

[0078] A particular example of a filter comprising five areas of interest is: F = + Max R paroiProximale − moy R CoucheMedianProximale + Max R ParoiDistale − moy R CoucheMedianDistale − moy R V a l l é e

[0079] Max is the maximum value of the signal over the area of ​​interest and avg() is the average of the signal over the area of ​​interest.

[0080] Another example implementation is to normalize the initial filter by the mean of the central region of interest or valley. F ′ = F moy ROI V a l l é e

[0081] In another embodiment, in the presence of similar anatomical objects, it is possible to increase the filter to consider extrinsic properties of the targeted object, for example if the radial artery has a vein on each side, the following form will make it possible to distinguish it: F " = F − α 6 f 6 ROI veineGauche − α 7 f 7 ROI veineDroite

[0082] In this embodiment, the filter is composed of seven zones.

[0083] Generally, the filter is defined over at least three consecutive time intervals by three respective functions each weighted by a coefficient, the coefficients associated with the first time interval and the third time interval being of the same sign and the coefficient associated with the second time interval being of opposite sign.

[0084] When the filter is defined over five time intervals, it is then defined over at least a first additional time interval located before the first time interval and a second additional time interval located after the third time interval, the coefficients associated with the two additional time intervals being of the same sign as the coefficient associated with the second time interval.

[0085] In other words, the filter coefficients corresponding to two consecutive intervals have opposite signs.

[0086] Generally speaking, alternating the signs of the filter coefficients over two consecutive intervals makes it possible to obtain a filtered signal which presents an extrema when the filter coincides temporally with the center of the interval delimited by two amplitude peaks of the initial signal, these peaks corresponding to the echoes of the signal on the walls of the blood vessel.

[0087] The implementation of the filter can be carried out in cascade, iteratively, for example by applying the filter to the entire signal during a first iteration then applying another filter to the result of the first filtering, centered on a reduced time zone, during the following iterations.

[0088] The filtering step 202 is applied to all the acquired ultrasonic signals corresponding to all the column vectors of the ultrasonic image of the figure 3a .

[0089] Alternatively or in addition, it is also possible to apply the same treatments to each line vector of the ultrasound image in order to identify the center of the artery according to the dimension parallel to the axis of the linear probe.

[0090] There figure 3b represents the result of applying the filter of the figure 4b for all signals (equivalent to the figure 4c for a signal).

[0091] There figure 3c represents the final result after zeroing the positive values. Artery 104 can be precisely identified on the image of the figure 3c .

[0092] The result of each filtering can be stored in memory and refreshed incrementally, keeping the optimum each time. It is also possible to partially store the optimal result of each signal, or to store all the filtering results entirely, depending on the available resources. Then, at step 203 of the method, we then search, among all the filtered signals, for the one that includes the highest amplitude extremum (in absolute value). The corresponding signal selected will be the one that corresponds to the path that passes through the center of the artery. Indeed, this signal must have the largest valley area (between the two walls of the artery) and therefore generate the largest negative contribution among all the signals. This is also reinforced by the wall which often has a maximum amplitude in the center.

[0093] In step 204, the center of the blood vessel is finally determined by recording the abscissa of the extremum measured on the selected signal.

[0094] This time abscissa t is then converted into distance d from the speed v of the ultrasonic signal in the medium: D = 1 2 V ⋅ t

[0095] A second embodiment of the invention is now described, which consists of subsequently determining the diameter of the blood vessel whose center has been detected.

[0096] This second embodiment is described in figure 5 .

[0097] It begins at step 501 by determining the center of the vessel using the method described previously in support of the figure 2 .

[0098] In step 502, the signal previously selected in step 203 is selected, which corresponds to the ultrasound path which passes through the center of the vessel.

[0099] There figure 6 shows an example of such a signal on which several extremum peaks have been identified.

[0100] In step 503, a threshold is applied to the signal in order to keep only the points corresponding to the extrema zones.

[0101] In step 504, the two points corresponding to the two extrema closest to the center of the vessel determined in step 501 are detected.

[0102] At step 505, the internal diameter of the vessel is deduced. Indeed, the two detected extrema correspond to the internal walls of the vessel. The internal diameter of the vessel is then equal à = 1 2 V ⋅ t 2 − t 1 , where t 2 , t 1 are the abscissas of the two extrema identified on the figure 6 .

[0103] In an optional step 506, the outer diameter of the vessel is further determined using the same relationship D = 1 2 V ⋅ t 4 − t 3 , where t 3 , t 4 are the abscissas of two other extrema exceeding the threshold of step 503 and located immediately after a low intensity zone which has low echogenicity and which is intermediate between the internal wall and the external wall of each of the proximal and distal sides. In other words, the extrema detected in step 506 are the extrema located immediately on either side of the first extrema detected in step 505 as shown in the figure 6 .

[0104] The different abscissas of the four extrema considered are identified on the figure 6 .

[0105] As explained previously, the method of the figure 5 can also be applied to the horizontal dimension of the ultrasound image, in order to determine the diameter of the vessel according to this dimension.

[0106] Although the invention has been described preferentially for medical applications consisting of characterizing a blood vessel and more precisely a radial artery, it can be applied more generally for any application involving ultrasound images of tubular-shaped objects. Références

[0107] [1] JH Gagan et al., “Automated Segmentation of Common Carotid Artery in Ultrasound Images,” in IEEE Access, vol. 10, pp. 58419-58430, 2022 [2] “System A feasibility study of a PMUT-based wearable sensor for the automatic monitoring of carotid artery parameters”, 2021 IEEE International Ultrasonics Symposium (IUS)

Claims

1. Method for characterizing a tubular-shaped object by ultrasound imaging, the method comprising the steps of: - Acquiring (201), by means of an ultrasound transducer, several ultrasound signals originating from the reflection of an ultrasound field emitted by the transducer on an area of interest in a cross-sectional plane of the object, for different positions of the transducer relative to said area, all of the ultrasound signals forming an ultrasound image of the area, - Choosing a dimension of the ultrasound image and for each signal corresponding to a vector of the image according to the chosen dimension, applying (202) a first predetermined filter to the signal, the filter being configured so as to transform a first signal comprising two extrema of the same sign into a second signal comprising an extremum of opposite sign located between the two extrema of the first signal, - Selecting (203), from all of the signals,the signal for which the filter result has the highest absolute value extremum and note the abscissa of this extremum, - Determine (204) the center of the object from the abscissa noted and the speed of the ultrasonic signal., 2. Method for characterizing a tubular-shaped object according to claim 1 in which the filter is applied to the envelope of the ultrasonic signal or to the absolute value of the ultrasonic signal.

3. Method for characterizing a tubular-shaped object according to claim 2, in which the filter is applied to the signal over a sliding window of predefined size depending on the size of the signal and / or a priori information on the dimension of the object, the filter being defined over at least three consecutive time intervals by three respective functions each weighted by a coefficient, the coefficients associated with two consecutive time intervals being of opposite signs.

4. Method for characterizing a tubular-shaped object according to claim 3, in which the dimension of the second time interval is chosen so as to be strictly less than the minimum diameter of the object to be characterized.

5. Method for characterizing a tubular-shaped object according to any one of claims 3 or 4 in which the filter is defined over at least two additional time intervals.

6. Method for characterizing a tubular-shaped object according to any one of claims 3 to 5 in which each of the functions is taken from: a sum, a maximum value, an average or a combination of these functions.

7. Method for characterizing a tubular-shaped object according to any one of the preceding claims further comprising the steps of: - Selecting (502) the acquired ultrasonic signal for which the center of the object has been determined, - Applying (503) a predetermined threshold to said selected ultrasonic signal, - Detecting (504) at least two extrema of said signal greater than the threshold, - Selecting the pair of extrema, comprising a first extremum and a second extremum, closest to the center of the object and located on either side of the center of the object, recording their respective time abscissas and deducing (505) the internal diameter of the object from the difference between the two abscissas and the speed of the ultrasonic signal.

8. Method for characterizing a tubular-shaped object according to claim 7 comprising the steps of: - Selecting a third extremum greater than the threshold and located immediately before the first extremum, - Selecting a fourth extremum greater than the threshold and located immediately after the second extremum, - Recording the time abscissas of the third extremum and the fourth extremum and deducing (506) the external diameter of the object from the difference between the two abscissas and the speed of the ultrasonic signal.

9. Method of characterizing a tubular-shaped object according to any one of the preceding claims in which the object is a blood vessel, for example an artery.

10. Method for characterizing a blood vessel according to claim 9 in which the step of acquiring (201) several ultrasound signals comprises the sub-steps of: - Positioning a transducer comprising several aligned elements, on an area of the skin so as to image a cross-section of the blood vessel, - Carrying out several successive ultrasound acquisitions from different emission points located on the alignment axis of the elements, each ultrasound emission being carried out in a direction substantially perpendicular to the alignment axis.

11. An ultrasound imaging device comprising an ultrasound transducer (101) and a processing unit configured to carry out the steps of the method according to any one of the preceding claims.

Images